Search

Your search keyword '"Finzi, A."' showing total 9,791 results
Start Over Author "Finzi, A."
9,791 results on '"Finzi, A."'

1. Searching for Efficient Linear Layers over a Continuous Space of Structured Matrices

Author

Potapczynski, Andres, Qiu, Shikai, Finzi, Marc, Ferri, Christopher, Chen, Zixi, Goldblum, Micah, Bruss, Bayan, De Sa, Christopher, and Wilson, Andrew Gordon

Subjects
Computer Science - Machine Learning, Statistics - Machine Learning
Abstract

Dense linear layers are the dominant computational bottleneck in large neural networks, presenting a critical need for more efficient alternatives. Previous efforts focused on a small number of hand-crafted structured matrices and neglected to investigate whether these structures can surpass dense layers in terms of compute-optimal scaling laws when both the model size and training examples are optimally allocated. In this work, we present a unifying framework that enables searching among all linear operators expressible via an Einstein summation. This framework encompasses many previously proposed structures, such as low-rank, Kronecker, Tensor-Train, Block Tensor-Train (BTT), and Monarch, along with many novel structures. To analyze the framework, we develop a taxonomy of all such operators based on their computational and algebraic properties and show that differences in the compute-optimal scaling laws are mostly governed by a small number of variables that we introduce. Namely, a small $\omega$ (which measures parameter sharing) and large $\psi$ (which measures the rank) reliably led to better scaling laws. Guided by the insight that full-rank structures that maximize parameters per unit of compute perform the best, we propose BTT-MoE, a novel Mixture-of-Experts (MoE) architecture obtained by sparsifying computation in the BTT structure. In contrast to the standard sparse MoE for each entire feed-forward network, BTT-MoE learns an MoE in every single linear layer of the model, including the projection matrices in the attention blocks. We find BTT-MoE provides a substantial compute-efficiency gain over dense layers and standard MoE., Comment: NeurIPS 2024. Code available at https://github.com/AndPotap/einsum-search

Published
2024

2. Unlocking Tokens as Data Points for Generalization Bounds on Larger Language Models

Author

Lotfi, Sanae, Kuang, Yilun, Amos, Brandon, Goldblum, Micah, Finzi, Marc, and Wilson, Andrew Gordon

Subjects
Statistics - Machine Learning, Computer Science - Machine Learning
Abstract

Large language models (LLMs) with billions of parameters excel at predicting the next token in a sequence. Recent work computes non-vacuous compression-based generalization bounds for LLMs, but these bounds are vacuous for large models at the billion-parameter scale. Moreover, these bounds are obtained through restrictive compression techniques, bounding compressed models that generate low-quality text. Additionally, the tightness of these existing bounds depends on the number of IID documents in a training set rather than the much larger number of non-IID constituent tokens, leaving untapped potential for tighter bounds. In this work, we instead use properties of martingales to derive generalization bounds that benefit from the vast number of tokens in LLM training sets. Since a dataset contains far more tokens than documents, our generalization bounds not only tolerate but actually benefit from far less restrictive compression schemes. With Monarch matrices, Kronecker factorizations, and post-training quantization, we achieve non-vacuous generalization bounds for LLMs as large as LLaMA2-70B. Unlike previous approaches, our work achieves the first non-vacuous bounds for models that are deployed in practice and generate high-quality text.

Published
2024

3. Compute Better Spent: Replacing Dense Layers with Structured Matrices

Author

Qiu, Shikai, Potapczynski, Andres, Finzi, Marc, Goldblum, Micah, and Wilson, Andrew Gordon

Subjects
Computer Science - Machine Learning
Abstract

Dense linear layers are the dominant computational bottleneck in foundation models. Identifying more efficient alternatives to dense matrices has enormous potential for building more compute-efficient models, as exemplified by the success of convolutional networks in the image domain. In this work, we systematically explore structured matrices as replacements for dense matrices. We show that different structures often require drastically different initialization scales and learning rates, which are crucial to performance, especially as models scale. Using insights from the Maximal Update Parameterization, we determine the optimal scaling for initialization and learning rates of these unconventional layers. Finally, we measure the scaling laws of different structures to compare how quickly their performance improves with compute. We propose a novel matrix family containing Monarch matrices, the Block Tensor-Train (BTT), which we show performs better than dense matrices for the same compute on multiple tasks. On CIFAR-10/100 with augmentation, BTT achieves exponentially lower training loss than dense when training MLPs and ViTs. BTT matches dense ViT-S/32 performance on ImageNet-1k with 3.8 times less compute and is more efficient than dense for training small GPT-2 language models., Comment: ICML 24. Code available at https://github.com/shikaiqiu/compute-better-spent

Published
2024

5. How to optimize neuroscience data utilization and experiment design for advancing primate visual and linguistic brain models?

Author

Tuckute, Greta, Finzi, Dawn, Margalit, Eshed, Zylberberg, Joel, Chung, SueYeon, Fyshe, Alona, Fedorenko, Evelina, Kriegeskorte, Nikolaus, Yates, Jacob, Grill-Spector, Kalanit, and Kar, Kohitij

Subjects
Quantitative Biology - Neurons and Cognition
Abstract

In recent years, neuroscience has made significant progress in building large-scale artificial neural network (ANN) models of brain activity and behavior. However, there is no consensus on the most efficient ways to collect data and design experiments to develop the next generation of models. This article explores the controversial opinions that have emerged on this topic in the domain of vision and language. Specifically, we address two critical points. First, we weigh the pros and cons of using qualitative insights from empirical results versus raw experimental data to train models. Second, we consider model-free (intuition-based) versus model-based approaches for data collection, specifically experimental design and stimulus selection, for optimal model development. Finally, we consider the challenges of developing a synergistic approach to experimental design and model building, including encouraging data and model sharing and the implications of iterative additions to existing models. The goal of the paper is to discuss decision points and propose directions for both experimenters and model developers in the quest to understand the brain.

Published
2024

6. Sustained IFN signaling is associated with delayed development of SARS-CoV-2-specific immunity.

Author

Brunet-Ratnasingham, Elsa, Morin, Sacha, Randolph, Haley, Labrecque, Marjorie, Bélair, Justin, Lima-Barbosa, Raphaël, Pagliuzza, Amélie, Marchitto, Lorie, Hultström, Michael, Niessl, Julia, Cloutier, Rose, Sreng Flores, Alina, Brassard, Nathalie, Benlarbi, Mehdi, Prévost, Jérémie, Ding, Shilei, Anand, Sai, Sannier, Gérémy, Marks, Amanda, Wågsäter, Dick, Bareke, Eric, Zeberg, Hugo, Lipcsey, Miklos, Frithiof, Robert, Larsson, Anders, Zhou, Sirui, Nakanishi, Tomoko, Morrison, David, Vezina, Dani, Bourassa, Catherine, Gendron-Lepage, Gabrielle, Medjahed, Halima, Point, Floriane, Richard, Jonathan, Larochelle, Catherine, Prat, Alexandre, Cunningham, Janet, Arbour, Nathalie, Durand, Madeleine, Richards, J, Moon, Kevin, Chomont, Nicolas, Finzi, Andrés, Tétreault, Martine, Barreiro, Luis, Wolf, Guy, and Kaufmann, Daniel

Subjects
Humans, COVID-19, SARS-CoV-2, Antibodies, Viral, Signal Transduction, Interferons, Female, Male, Middle Aged, Immunoglobulin G, CD4-Positive T-Lymphocytes, Aged, Adult, Spike Glycoprotein, Coronavirus
Abstract

Plasma RNAemia, delayed antibody responses and inflammation predict COVID-19 outcomes, but the mechanisms underlying these immunovirological patterns are poorly understood. We profile 782 longitudinal plasma samples from 318 hospitalized patients with COVID-19. Integrated analysis using k-means reveals four patient clusters in a discovery cohort: mechanically ventilated critically-ill cases are subdivided into good prognosis and high-fatality clusters (reproduced in a validation cohort), while non-critical survivors segregate into high and low early antibody responders. Only the high-fatality cluster is enriched for transcriptomic signatures associated with COVID-19 severity, and each cluster has distinct RBD-specific antibody elicitation kinetics. Both critical and non-critical clusters with delayed antibody responses exhibit sustained IFN signatures, which negatively correlate with contemporaneous RBD-specific IgG levels and absolute SARS-CoV-2-specific B and CD4+ T cell frequencies. These data suggest that the Interferon paradox previously described in murine LCMV models is operative in COVID-19, with excessive IFN signaling delaying development of adaptive virus-specific immunity.

Published
2024

8. Non-Vacuous Generalization Bounds for Large Language Models

Author

Lotfi, Sanae, Finzi, Marc, Kuang, Yilun, Rudner, Tim G. J., Goldblum, Micah, and Wilson, Andrew Gordon

Subjects
Statistics - Machine Learning, Computer Science - Machine Learning
Abstract

Modern language models can contain billions of parameters, raising the question of whether they can generalize beyond the training data or simply parrot their training corpora. We provide the first non-vacuous generalization bounds for pretrained large language models (LLMs), indicating that language models are capable of discovering regularities that generalize to unseen data. In particular, we derive a compression bound that is valid for the unbounded log-likelihood loss using prediction smoothing, and we extend the bound to handle subsampling, accelerating bound computation by orders of magnitude on massive datasets. To achieve the extreme level of compression required for non-vacuous bounds, we devise SubLoRA, a simple low-dimensional nonlinear parameterization that leads to non-vacuous generalization bounds for models with nearly a billion parameters. Finally, we use our bounds to understand LLM generalization and find that larger models have better generalization bounds and are more compressible than smaller models., Comment: ICML 2024

Published
2023

9. Large Language Models Are Zero-Shot Time Series Forecasters

Author

Gruver, Nate, Finzi, Marc, Qiu, Shikai, and Wilson, Andrew Gordon

Subjects
Computer Science - Machine Learning
Abstract

By encoding time series as a string of numerical digits, we can frame time series forecasting as next-token prediction in text. Developing this approach, we find that large language models (LLMs) such as GPT-3 and LLaMA-2 can surprisingly zero-shot extrapolate time series at a level comparable to or exceeding the performance of purpose-built time series models trained on the downstream tasks. To facilitate this performance, we propose procedures for effectively tokenizing time series data and converting discrete distributions over tokens into highly flexible densities over continuous values. We argue the success of LLMs for time series stems from their ability to naturally represent multimodal distributions, in conjunction with biases for simplicity, and repetition, which align with the salient features in many time series, such as repeated seasonal trends. We also show how LLMs can naturally handle missing data without imputation through non-numerical text, accommodate textual side information, and answer questions to help explain predictions. While we find that increasing model size generally improves performance on time series, we show GPT-4 can perform worse than GPT-3 because of how it tokenizes numbers, and poor uncertainty calibration, which is likely the result of alignment interventions such as RLHF., Comment: NeurIPS 2023. Code available at: https://github.com/ngruver/llmtime

Published
2023

13. CoLA: Exploiting Compositional Structure for Automatic and Efficient Numerical Linear Algebra

Author

Potapczynski, Andres, Finzi, Marc, Pleiss, Geoff, and Wilson, Andrew Gordon

Subjects
Computer Science - Machine Learning, Mathematics - Numerical Analysis, Statistics - Machine Learning
Abstract

Many areas of machine learning and science involve large linear algebra problems, such as eigendecompositions, solving linear systems, computing matrix exponentials, and trace estimation. The matrices involved often have Kronecker, convolutional, block diagonal, sum, or product structure. In this paper, we propose a simple but general framework for large-scale linear algebra problems in machine learning, named CoLA (Compositional Linear Algebra). By combining a linear operator abstraction with compositional dispatch rules, CoLA automatically constructs memory and runtime efficient numerical algorithms. Moreover, CoLA provides memory efficient automatic differentiation, low precision computation, and GPU acceleration in both JAX and PyTorch, while also accommodating new objects, operations, and rules in downstream packages via multiple dispatch. CoLA can accelerate many algebraic operations, while making it easy to prototype matrix structures and algorithms, providing an appealing drop-in tool for virtually any computational effort that requires linear algebra. We showcase its efficacy across a broad range of applications, including partial differential equations, Gaussian processes, equivariant model construction, and unsupervised learning., Comment: Code available at https://github.com/wilson-labs/cola

Published
2023

16. User-defined Event Sampling and Uncertainty Quantification in Diffusion Models for Physical Dynamical Systems

Author

Finzi, Marc, Boral, Anudhyan, Wilson, Andrew Gordon, Sha, Fei, and Zepeda-Núñez, Leonardo

Subjects
Computer Science - Machine Learning, Computer Science - Artificial Intelligence
Abstract

Diffusion models are a class of probabilistic generative models that have been widely used as a prior for image processing tasks like text conditional generation and inpainting. We demonstrate that these models can be adapted to make predictions and provide uncertainty quantification for chaotic dynamical systems. In these applications, diffusion models can implicitly represent knowledge about outliers and extreme events; however, querying that knowledge through conditional sampling or measuring probabilities is surprisingly difficult. Existing methods for conditional sampling at inference time seek mainly to enforce the constraints, which is insufficient to match the statistics of the distribution or compute the probability of the chosen events. To achieve these ends, optimally one would use the conditional score function, but its computation is typically intractable. In this work, we develop a probabilistic approximation scheme for the conditional score function which provably converges to the true distribution as the noise level decreases. With this scheme we are able to sample conditionally on nonlinear userdefined events at inference time, and matches data statistics even when sampling from the tails of the distribution., Comment: ICML 2023 Conference

Published
2023

17. A Stable and Scalable Method for Solving Initial Value PDEs with Neural Networks

Author

Finzi, Marc, Potapczynski, Andres, Choptuik, Matthew, and Wilson, Andrew Gordon

Subjects
Computer Science - Machine Learning, Mathematics - Numerical Analysis, Statistics - Machine Learning
Abstract

Unlike conventional grid and mesh based methods for solving partial differential equations (PDEs), neural networks have the potential to break the curse of dimensionality, providing approximate solutions to problems where using classical solvers is difficult or impossible. While global minimization of the PDE residual over the network parameters works well for boundary value problems, catastrophic forgetting impairs the applicability of this approach to initial value problems (IVPs). In an alternative local-in-time approach, the optimization problem can be converted into an ordinary differential equation (ODE) on the network parameters and the solution propagated forward in time; however, we demonstrate that current methods based on this approach suffer from two key issues. First, following the ODE produces an uncontrolled growth in the conditioning of the problem, ultimately leading to unacceptably large numerical errors. Second, as the ODE methods scale cubically with the number of model parameters, they are restricted to small neural networks, significantly limiting their ability to represent intricate PDE initial conditions and solutions. Building on these insights, we develop Neural IVP, an ODE based IVP solver which prevents the network from getting ill-conditioned and runs in time linear in the number of parameters, enabling us to evolve the dynamics of challenging PDEs with neural networks., Comment: ICLR 2023. Code available at https://github.com/mfinzi/neural-ivp

Published
2023

18. The No Free Lunch Theorem, Kolmogorov Complexity, and the Role of Inductive Biases in Machine Learning

Author

Goldblum, Micah, Finzi, Marc, Rowan, Keefer, and Wilson, Andrew Gordon

Subjects
Computer Science - Machine Learning, Statistics - Machine Learning
Abstract

No free lunch theorems for supervised learning state that no learner can solve all problems or that all learners achieve exactly the same accuracy on average over a uniform distribution on learning problems. Accordingly, these theorems are often referenced in support of the notion that individual problems require specially tailored inductive biases. While virtually all uniformly sampled datasets have high complexity, real-world problems disproportionately generate low-complexity data, and we argue that neural network models share this same preference, formalized using Kolmogorov complexity. Notably, we show that architectures designed for a particular domain, such as computer vision, can compress datasets on a variety of seemingly unrelated domains. Our experiments show that pre-trained and even randomly initialized language models prefer to generate low-complexity sequences. Whereas no free lunch theorems seemingly indicate that individual problems require specialized learners, we explain how tasks that often require human intervention such as picking an appropriately sized model when labeled data is scarce or plentiful can be automated into a single learning algorithm. These observations justify the trend in deep learning of unifying seemingly disparate problems with an increasingly small set of machine learning models., Comment: Published at the International Conference on Machine Learning (ICML) 2024

Published
2023

19. Color Doppler Imaging, Endothelin-1, Corneal Biomechanics and Scleral Rigidity in Asymmetric Age-Related Macular Degeneration

Author

Finzi A, Ottoboni S, Cellini M, Corcioni B, Gaudiano C, and Fontana L

Subjects
amd, endothelin-1, corneal biomechanics, scleral rigidity, hemodynamics, Ophthalmology, RE1-994
Abstract

Alessandro Finzi,1 Simone Ottoboni,1 Mauro Cellini,1 Beniamino Corcioni,2 Caterina Gaudiano,2 Luigi Fontana1 1Ophthalmology Unit, Department of Experimental, Diagnostic and Specialty Medicine, IRCCS Azienda Ospedaliero-Universitaria di Bologna, Bologna, Italy; 2Department of Radiology, IRCCS Azienda Ospedaliero-Universitaria di Bologna, Bologna, ItalyCorrespondence: Alessandro Finzi, Ophthalmology Unit, Department of Experimental, Diagnostic and Specialty Medicine, IRCCS Azienda Ospedaliero-Universitaria di Bologna, Via Pelagio Palagi 9, Bologna, 40138, Italy, Tel +39 051 2142831, Email finzialessandro@gmail.comPurpose: Age-related macular degeneration (AMD) presents a multifaceted etiopathogenesis involving ischemic, inflammatory, and genetic components. This study investigates the correlation between ocular hemodynamics, scleral rigidity (SR), and plasma endothelin-1 (ET1) levels in treatment-naive patients with asymmetrical AMD.Patients and Methods: This study included 20 treatment-naive patients (12 females and 8 males) with an average age of 76.4 ± 3.7 years, who presented with AMD with neovascular membrane formation (nAMD) in one eye, and intermediate grade 2 AMD (iAMD) in the other eye. The control group consisted of 20 healthy subjects (13 females and 7 males) with a mean age of 74.7 ± 3.9 years. All patients and healthy controls underwent color Doppler imaging (i) of the ophthalmic artery (OA), short posterior ciliary arteries (SPCAs), and central retinal artery (CRA); Plasma ET-1 levels were measured for all patients and healthy subjects. Corneal biomechanics were assessed using an Ocular Response Analyzer and two indices were obtained: corneal hysteresis (CH) and corneal resistance factor (CRF).Results: Results showed reduced blood flow velocities and increased resistance indices in AMD eyes, particularly affecting the short posterior ciliary arteries. According to mechanical theory, ARMD eyes exhibited elevated scleral rigidity and corneal resistance factor compared to controls, with a notable rise in SR in neovascular AMD (nAMD) eyes. As per the chronic subacute inflammation theory, plasma ET-1 levels were significantly higher in AMD patients, correlating with abnormal SPCAs blood flow and increased resistance indices.Conclusion: Findings suggest a multifactorial etiology of AMD involving an increase of ET-1 plasma levels with biomechanic damages of corneal and scleral tissue in nAMD.Keywords: AMD, endothelin-1, corneal biomechanics, scleral rigidity, hemodynamics

Published
2024

20. Force and the α-C-terminal domains bias RNA polymerase recycling

Author

Jin Qian, Bing Wang, Irina Artsimovitch, David Dunlap, and Laura Finzi

Subjects
Science
Abstract

Abstract After an RNA polymerase reaches a terminator, instead of dissociating from the template, it may diffuse along the DNA and recommence RNA synthesis from the previous or a different promoter. Magnetic tweezers were used to monitor such secondary transcription and determine the effects of low forces assisting or opposing translocation, protein roadblocks, and transcription factors. Remarkably, up to 50% of Escherichia coli (E. coli) RNA polymerases diffused along the DNA after termination. Force biased the direction of diffusion (sliding) and the velocity increased rapidly with force up to 0.7 pN and much more slowly thereafter. Sigma factor 70 (σ 70) likely remained associated with the DNA promoting sliding and enabling re-initiation from promoters in either orientation. However, deletions of the α-C-terminal domains severely limited the ability of RNAP to turn around between successive rounds of transcription. The addition of elongation factor NusG, which competes with σ 70 for binding to RNAP, limited additional rounds of transcription. Surprisingly, sliding RNA polymerases blocked by a DNA-bound lac repressor could slowly re-initiate transcription and were not affected by NusG, suggesting a σ-independent pathway. Low forces effectively biased promoter selection suggesting a prominent role for topological entanglements that affect RNA polymerase translocation.

Published
2024
Full Text
View/download PDF

21. Pneumatic retinopexy for primary rhegmatogenous retinal detachment: from a clinical trial to the real-life experience

Author

Danilo Iannetta, Nicola Valsecchi, Alessandro Finzi, Rodolfo Mastropasqua, Rajeev H. Muni, and Luigi Fontana

Subjects
Pneumatic Retinopexy, Rhegmatogenous Retinal detachment, Real-world outcomes, Surgical success, Complications, Ophthalmology, RE1-994
Abstract

Abstract Background To report real-world outcomes of patients with primary Reghmatogenous Retinal Detachment (RRD) treated with Pneumatic Retinopexy (PnR) according to the indications of the Pneumatic Retinopexy versus Vitrectomy for management of Primary Rhegmatogenous Retinal Detachment Outcomes Randomized Trial (PIVOT) trial. Methods Multicenter, retrospective study. Patients treated with PnR for RRD between 2021 and 2023 and a follow-up of at least 6 months were included. Single-procedure anatomical success, final anatomical success, complications, causes of failures, best corrected visual acuity (BCVA) after surgery, and the vision-related quality of life using the 25-Item National Eye Institute Visual Function Questionnaire (NEI VFQ-25) were reported. Results A total of 76 eyes of 76 patients were included. Mean age was 60 ± 8.1 years. Primary anatomic reattachment was achieved by 84.3% of patients and final anatomical reattachment after pars plana vitrectomy was obtained in 100% of patients. BCVA improved from 0.32 (20/40) to 0.04 (20/20) logMar (p

Published
2024
Full Text
View/download PDF

23. CD4 downregulation precedes Env expression and protects HIV-1-infected cells from ADCC mediated by non-neutralizing antibodies

Author

Jonathan Richard, Gérémy Sannier, Li Zhu, Jérémie Prévost, Lorie Marchitto, Mehdi Benlarbi, Guillaume Beaudoin-Bussières, Hongil Kim, Yaping Sun, Debashree Chatterjee, Halima Medjahed, Catherine Bourassa, Gloria-Gabrielle Delgado, Mathieu Dubé, Frank Kirchhoff, Beatrice H. Hahn, Priti Kumar, Daniel E. Kaufmann, and Andrés Finzi

Subjects
HIV-1, ADCC, Env, nnAbs, bNAbs, hu-mice, Microbiology, QR1-502
Abstract

ABSTRACT HIV-1 envelope glycoprotein (Env) conformation substantially impacts antibody-dependent cellular cytotoxicity (ADCC). Envs from primary HIV-1 isolates adopt a prefusion “closed” conformation, which is targeted by broadly neutralizing antibodies (bnAbs). CD4 binding drives Env into more “open” conformations, which are recognized by non-neutralizing Abs (nnAbs). To better understand Env–Ab and Env–CD4 interaction in CD4+ T cells infected with HIV-1, we simultaneously measured antibody binding and HIV-1 mRNA expression using multiparametric flow cytometry and RNA flow fluorescent in situ hybridization (FISH) techniques. We observed that env mRNA is almost exclusively expressed by HIV-1 productively infected cells that already downmodulated CD4. This suggests that CD4 downmodulation precedes env mRNA expression. Consequently, productively infected cells express “closed” Envs on their surface, which renders them resistant to nnAbs. Cells recognized by nnAbs were all env mRNA negative, indicating Ab binding through shed gp120 or virions attached to their surface. Consistent with these findings, treatment of HIV-1-infected humanized mice with the ADCC-mediating nnAb A32 failed to lower viral replication or reduce the size of the viral reservoir. These findings confirm the resistance of productively infected CD4+ T cells to nnAbs-mediated ADCC and question the rationale of immunotherapy approaches using this strategy.IMPORTANCEAntibody-dependent cellular cytotoxicity (ADCC) represents an effective immune response for clearing virally infected cells, making ADCC-mediating antibodies promising therapeutic candidates for HIV-1 cure strategies. Broadly neutralizing antibodies (bNAbs) target epitopes present on the native "closed" envelope glycoprotein (Env), while non-neutralizing antibodies (nnAbs) recognize epitopes exposed upon Env–CD4 interaction. Here, we provide evidence that env mRNA is predominantly expressed by productively infected cells that have already downmodulated cell-surface CD4. This indicates that CD4 downmodulation by HIV-1 precedes Env expression, making productively infected cells resistant to ADCC mediated by nnAbs but sensitive to those mediated by bnAbs. These findings offer critical insights for the development of immunotherapy-based strategies aimed at targeting and eliminating productively infected cells in people living with HIV.

Published
2024
Full Text
View/download PDF

25. Combining Hierarchical MILP-MPC and Artificial Potential Fields for Multi-robot Priority-Based Sanitization of Railway Stations

Author

Caccavale, Riccardo, Ermini, Mirko, Fedeli, Eugenio, Finzi, Alberto, Garone, Emanuele, Lippiello, Vincenzo, Tavano, Fabrizio, Siciliano, Bruno, Series Editor, Khatib, Oussama, Series Editor, Antonelli, Gianluca, Advisory Editor, Fox, Dieter, Advisory Editor, Harada, Kensuke, Advisory Editor, Hsieh, M. Ani, Advisory Editor, Kröger, Torsten, Advisory Editor, Kulic, Dana, Advisory Editor, Park, Jaeheung, Advisory Editor, Bourgeois, Julien, editor, Paik, Jamie, editor, Piranda, Benoît, editor, Werfel, Justin, editor, Hauert, Sabine, editor, Pierson, Alyssa, editor, Hamann, Heiko, editor, Lam, Tin Lun, editor, Matsuno, Fumitoshi, editor, Mehr, Negar, editor, and Makhoul, Abdallah, editor

Published
2024
Full Text
View/download PDF

26. Sunspot periodicity

Author

Vita-Finzi, Claudio

Subjects
Astrophysics - Solar and Stellar Astrophysics, Physics - Geophysics, Physics - History and Philosophy of Physics
Abstract

The Schwabe (~11 yr) value for the annual sunspot number is sometimes uncritically applied to other measures of solar activity, direct and indirect, including the 10.7 cm radio flux, the inflow of galactic cosmic rays, solar flare frequency, terrestrial weather, and components of space climate, with the risk of a resulting loss of information. The ruling (Babcock) hypothesis and its derivatives link the sunspot cycle to dynamo processes mediated by differential solar rotation, but despite 60 years of observation and analysis the ~11 yr periodicity remains difficult to model; the possible contribution of planetary dynamics is undergoing a revival. The various solar sequences that genuinely display an ~11 yr cycle stand to benefit from an understanding of its periodicity that goes beyond statistical kinship. The outcome could ironically prompt the demotion of sunspots from their dominant historical role in favour of other possible indicators of solar cyclicity, such as the solar wind flux and its isotopic signatures, even if they are less accessible., Comment: 28 pages

Published
2022

27. PAC-Bayes Compression Bounds So Tight That They Can Explain Generalization

Author

Lotfi, Sanae, Finzi, Marc, Kapoor, Sanyam, Potapczynski, Andres, Goldblum, Micah, and Wilson, Andrew Gordon

Subjects
Computer Science - Machine Learning, Statistics - Machine Learning
Abstract

While there has been progress in developing non-vacuous generalization bounds for deep neural networks, these bounds tend to be uninformative about why deep learning works. In this paper, we develop a compression approach based on quantizing neural network parameters in a linear subspace, profoundly improving on previous results to provide state-of-the-art generalization bounds on a variety of tasks, including transfer learning. We use these tight bounds to better understand the role of model size, equivariance, and the implicit biases of optimization, for generalization in deep learning. Notably, we find large models can be compressed to a much greater extent than previously known, encapsulating Occam's razor. We also argue for data-independent bounds in explaining generalization., Comment: NeurIPS 2022. Code is available at https://github.com/activatedgeek/tight-pac-bayes

Published
2022

28. The Lie Derivative for Measuring Learned Equivariance

Author

Gruver, Nate, Finzi, Marc, Goldblum, Micah, and Wilson, Andrew Gordon

Subjects
Computer Science - Machine Learning, Computer Science - Artificial Intelligence, Computer Science - Computer Vision and Pattern Recognition, Statistics - Machine Learning
Abstract

Equivariance guarantees that a model's predictions capture key symmetries in data. When an image is translated or rotated, an equivariant model's representation of that image will translate or rotate accordingly. The success of convolutional neural networks has historically been tied to translation equivariance directly encoded in their architecture. The rising success of vision transformers, which have no explicit architectural bias towards equivariance, challenges this narrative and suggests that augmentations and training data might also play a significant role in their performance. In order to better understand the role of equivariance in recent vision models, we introduce the Lie derivative, a method for measuring equivariance with strong mathematical foundations and minimal hyperparameters. Using the Lie derivative, we study the equivariance properties of hundreds of pretrained models, spanning CNNs, transformers, and Mixer architectures. The scale of our analysis allows us to separate the impact of architecture from other factors like model size or training method. Surprisingly, we find that many violations of equivariance can be linked to spatial aliasing in ubiquitous network layers, such as pointwise non-linearities, and that as models get larger and more accurate they tend to display more equivariance, regardless of architecture. For example, transformers can be more equivariant than convolutional neural networks after training., Comment: ICLR 2023. Code available at: https://github.com/ngruver/lie-deriv

Published
2022

30. Treatment Patterns and Clinical Outcomes of Patients with Moderate to Severe Acute Graft-Versus-Host Disease: A Multicenter Chart Review Study

Author

David Michonneau, Raynier Devillier, Mikko Keränen, Marie Thérèse Rubio, Malin Nicklasson, Hélène Labussière-Wallet, Martin Carre, Anne Huynh, Elisabet Viayna, Montserrat Roset, Jonathan Finzi, Minja Pfeiffer, Daniel Thunström, Núria Lara, Lorenzo Sabatelli, Patrice Chevallier, and Maija Itälä-Remes

Subjects
hematopoietic stem cell transplantation, Europe, retrospective, mortality, real-world, Diseases of the blood and blood-forming organs, RC633-647.5
Abstract

Acute graft-versus-host disease (aGVHD) remains a barrier to successful allogeneic hematopoietic stem cell transplantation (HSCT) outcomes. Contemporary comprehensive analyses of real-world clinical outcomes among patients who develop aGVHD post-HSCT are needed to better understand the unmet needs of this patient population. This multicenter, retrospective chart review describes treatment patterns and clinical outcomes among patients (≥18 years old) from Finland, Sweden, and France who developed grades II–IV aGVHD after their first HSCT (January 2016–June 2017). From 13 participating centers, 151 patients were included. The median (Q1, Q3) age at HSCT was 56 (45, 62) years old. One line of aGVHD treatment was received by 47.7%, and the most common first-line treatment was methylprednisolone (alone or in a combination regimen, 74.2%; monotherapy, 25.8%). Among patients treated with methylprednisolone, 79.5% achieved a complete or partial response. The median (Q1, Q3) number of treatment lines was 2.0 (1.0, 3.0). The median (Q1, Q3) time to obtain an aGVHD diagnosis from transplant was 29.5 (21.0, 44.0) days, and 14.5 (7.0, 34.0) days to achieve the best response for 110 evaluable patients. At 6 and 12 months, 53.6% and 49.0%, respectively, achieved a complete response. Chronic GVHD occurred in 37.7% of patients, and aGVHD reoccurred in 26.5%. Following aGVHD diagnosis, mortality rates were 30.0% at 6 months and 37.3% at 12 months. Findings from this study demonstrate a continuing unmet need for new therapies that control aGVHD and improve mortality.

Published
2024
Full Text
View/download PDF

31. Sustained IFN signaling is associated with delayed development of SARS-CoV-2-specific immunity

Author

Elsa Brunet-Ratnasingham, Sacha Morin, Haley E. Randolph, Marjorie Labrecque, Justin Bélair, Raphaël Lima-Barbosa, Amélie Pagliuzza, Lorie Marchitto, Michael Hultström, Julia Niessl, Rose Cloutier, Alina M. Sreng Flores, Nathalie Brassard, Mehdi Benlarbi, Jérémie Prévost, Shilei Ding, Sai Priya Anand, Gérémy Sannier, Amanda Marks, Dick Wågsäter, Eric Bareke, Hugo Zeberg, Miklos Lipcsey, Robert Frithiof, Anders Larsson, Sirui Zhou, Tomoko Nakanishi, David Morrison, Dani Vezina, Catherine Bourassa, Gabrielle Gendron-Lepage, Halima Medjahed, Floriane Point, Jonathan Richard, Catherine Larochelle, Alexandre Prat, Janet L. Cunningham, Nathalie Arbour, Madeleine Durand, J. Brent Richards, Kevin Moon, Nicolas Chomont, Andrés Finzi, Martine Tétreault, Luis Barreiro, Guy Wolf, and Daniel E. Kaufmann

Subjects
Science
Abstract

Abstract Plasma RNAemia, delayed antibody responses and inflammation predict COVID-19 outcomes, but the mechanisms underlying these immunovirological patterns are poorly understood. We profile 782 longitudinal plasma samples from 318 hospitalized patients with COVID-19. Integrated analysis using k-means reveals four patient clusters in a discovery cohort: mechanically ventilated critically-ill cases are subdivided into good prognosis and high-fatality clusters (reproduced in a validation cohort), while non-critical survivors segregate into high and low early antibody responders. Only the high-fatality cluster is enriched for transcriptomic signatures associated with COVID-19 severity, and each cluster has distinct RBD-specific antibody elicitation kinetics. Both critical and non-critical clusters with delayed antibody responses exhibit sustained IFN signatures, which negatively correlate with contemporaneous RBD-specific IgG levels and absolute SARS-CoV-2-specific B and CD4+ T cell frequencies. These data suggest that the “Interferon paradox” previously described in murine LCMV models is operative in COVID-19, with excessive IFN signaling delaying development of adaptive virus-specific immunity.

Published
2024
Full Text
View/download PDF

32. Reciprocating RNA Polymerase batters through roadblocks

Author

Jin Qian, Allison Cartee, Wenxuan Xu, Yan Yan, Bing Wang, Irina Artsimovitch, David Dunlap, and Laura Finzi

Subjects
Science
Abstract

Abstract RNA polymerases must transit through protein roadblocks to produce full-length transcripts. Here we report real-time measurements of Escherichia coli RNA polymerase passing through different barriers. As intuitively expected, assisting forces facilitated, and opposing forces hindered, RNA polymerase passage through lac repressor protein bound to natural binding sites. Force-dependent differences were significant at magnitudes as low as 0.2 pN and were abolished in the presence of the transcript cleavage factor GreA, which rescues backtracked RNA polymerase. In stark contrast, opposing forces promoted passage when the rate of RNA polymerase backtracking was comparable to, or faster than the rate of dissociation of the roadblock, particularly in the presence of GreA. Our experiments and simulations indicate that RNA polymerase may transit after roadblocks dissociate, or undergo cycles of backtracking, recovery, and ramming into roadblocks to pass through. We propose that such reciprocating motion also enables RNA polymerase to break protein-DNA contacts that hold RNA polymerase back during promoter escape and RNA chain elongation. This may facilitate productive transcription in vivo.

Published
2024
Full Text
View/download PDF

36. Metformin facilitates viral reservoir reactivation and their recognition by anti-HIV-1 envelope antibodies

Author

Augustine Fert, Jonathan Richard, Laurence Raymond Marchand, Delphine Planas, Jean-Pierre Routy, Nicolas Chomont, Andrés Finzi, and Petronela Ancuta

Subjects
Medication, Virology, Molecular microbiology, Science
Abstract

Summary: The mechanistic target of rapamycin (mTOR) positively regulates multiple steps of the HIV-1 replication cycle. We previously reported that a 12-week supplementation of antiretroviral therapy (ART) with metformin, an indirect mTOR inhibitor used in type-2 diabetes treatment, reduced mTOR activation and HIV transcription in colon-infiltrating CD4+ T cells, together with systemic inflammation in nondiabetic people with HIV-1 (PWH). Herein, we investigated the antiviral mechanisms of metformin. In a viral outgrowth assay performed with CD4+ T cells from ART-treated PWH, and upon infection in vitro with replication-competent and VSV-G-pseudotyped HIV-1, metformin decreased virion release, but increased the frequency of productively infected CD4lowHIV-p24+ T cells. These observations coincided with increased BST2/tetherin (HIV release inhibitor) and Bcl-2 (pro-survival factor) expression, and improved recognition of productively infected T cells by HIV-1 envelope antibodies. Thus, metformin exerts pleiotropic effects on post-integration steps of the HIV-1 replication cycle and may be used to accelerate viral reservoir decay in ART-treated PWH.

Published
2024
Full Text
View/download PDF

37. Temperature-dependent Spike-ACE2 interaction of Omicron subvariants is associated with viral transmission

Author

Mehdi Benlarbi, Shilei Ding, Étienne Bélanger, Alexandra Tauzin, Raphaël Poujol, Halima Medjahed, Omar El Ferri, Yuxia Bo, Catherine Bourassa, Julie Hussin, Judith Fafard, Marzena Pazgier, Inès Levade, Cameron Abrams, Marceline Côté, and Andrés Finzi

Subjects
SARS-CoV-2, Omicron subvariants, humoral responses, ACE2 binding, temperature, Microbiology, QR1-502
Abstract

ABSTRACT The continued evolution of severe acute respiratory syndrome 2 (SARS-CoV-2) requires persistent monitoring of its subvariants. Omicron subvariants are responsible for the vast majority of SARS-CoV-2 infections worldwide, with XBB and BA.2.86 sublineages representing more than 90% of circulating strains as of January 2024. To better understand parameters involved in viral transmission, we characterized the functional properties of Spike glycoproteins from BA.2.75, CH.1.1, DV.7.1, BA.4/5, BQ.1.1, XBB, XBB.1, XBB.1.16, XBB.1.5, FD.1.1, EG.5.1, HK.3, BA.2.86 and JN.1. We tested their capacity to evade plasma-mediated recognition and neutralization, binding to angiotensin-converting enzyme 2 (ACE2), their susceptibility to cold inactivation, Spike processing, as well as the impact of temperature on Spike-ACE2 interaction. We found that compared to the early wild-type (D614G) strain, most Omicron subvariants' Spike glycoproteins evolved to escape recognition and neutralization by plasma from individuals who received a fifth dose of bivalent (BA.1 or BA.4/5) mRNA vaccine and improve ACE2 binding, particularly at low temperatures. Moreover, BA.2.86 had the best affinity for ACE2 at all temperatures tested. We found that Omicron subvariants’ Spike processing is associated with their susceptibility to cold inactivation. Intriguingly, we found that Spike-ACE2 binding at low temperature was significantly associated with growth rates of Omicron subvariants in humans. Overall, we report that Spikes from newly emerged Omicron subvariants are relatively more stable and resistant to plasma-mediated neutralization, present improved affinity for ACE2 which is associated, particularly at low temperatures, with their growth rates.IMPORTANCEThe persistent evolution of SARS-CoV-2 gave rise to a wide range of variants harboring new mutations in their Spike glycoproteins. Several factors have been associated with viral transmission and fitness such as plasma-neutralization escape and ACE2 interaction. To better understand whether additional factors could be of importance in SARS-CoV-2 variants’ transmission, we characterize the functional properties of Spike glycoproteins from several Omicron subvariants. We found that the Spike glycoprotein of Omicron subvariants presents an improved escape from plasma-mediated recognition and neutralization, Spike processing, and ACE2 binding which was further improved at low temperature. Intriguingly, Spike-ACE2 interaction at low temperature is strongly associated with viral growth rate, as such, low temperatures could represent another parameter affecting viral transmission.

Published
2024
Full Text
View/download PDF

38. Nitrogen recovery from intensive livestock farms using a simplified ammonia stripping process

Author

Alberto Finzi, Omar Ferrari, Elisabetta Riva, and Giorgio Provolo

Subjects
digestate, pig and cattle slurry, nitrogen recovery, ammonium sulfate, pilot plant, Nutrition. Foods and food supply, TX341-641, Food processing and manufacture, TP368-456
Abstract

The ammonia (NH3) stripping process can recover nitrogen (N) from slurry and digestates as a mineral fertilizer, but it is currently expensive and difficult to manage at the farm level. Hence, a simple process is required. This study aimed to test a modular (based on farm N surplus) slow-release NH3 stripping process at a pilot plant scale. NH3 volatilization was promoted in a closed reactor, and then, the NH3 was removed by an air stream through the reactor headspace. The NH3-loaded air was purified in a scrubber, where NH3 reacted with sulfuric acid to form ammonium sulfate (AS). In total, 11 trials were conducted using pig slurry, dairy cattle slurry, and digestates: 7 trials were carried out with the reactors heated to 40°C, 2 trials were carried out at an ambient temperature, and the other 2 trials were carried out with the addition of sodium hydroxide (NaOH). To assess the technical–economic sustainability of the pilot plant, the total ammoniacal nitrogen (TAN) removal rate, electricity consumption, acid requirements, and AS quality and costs were evaluated. The pilot plant yielded TAN recovery amounts of 45% in 2 weeks with the reactors heated to 40°C, 64% in 1 week with NaOH addition, and 25% in 2 weeks at an ambient temperature. The N concentration in the AS solution reached 85.9 g kg−1, with an average value of 35.2 g kg−1. The electricity consumption, acid requirement, and operational costs in an optimized system were approximately 0.52 kWh kgN−1 recovered, 3.5 kg pure acid kgN−1 recovered, and 0.86 € kgN−1 recovered, respectively. Compared to other technologies, the simplified stripping process is slower but with similar removal efficiencies and lower energy consumption. Thus, this finding could be suitable for improving the N use in intensive livestock farms.

Published
2024
Full Text
View/download PDF

39. A Real-time Calculus Approach for Integrating Sporadic Events in Time-triggered Systems

Author

Finzi, Anaïs, Craciunas, Silviu S., and Boyer, Marc

Subjects
Computer Science - Networking and Internet Architecture, Electrical Engineering and Systems Science - Systems and Control
Abstract

In time-triggered systems, where the schedule table is predefined and statically configured at design time, sporadic event-triggered (ET) tasks are handled within specially dedicated slots or when time-triggered (TT) tasks finish their execution early. We introduce a new paradigm for synthesizing TT schedules that guarantee the correct temporal behavior of TT tasks and the schedulability of sporadic ET tasks with arbitrary deadlines. The approach first expresses a constraint for the TT task schedule in the form of a maximal affine envelope that guarantees that as long as the schedule generation respects this envelope, all sporadic ET tasks meet their deadline. The second step consists of modeling this envelope as a burst limiting constraint and building the TT schedule via simulating a modified Least-Laxity-First (LLF) scheduler. Using this novel technique, we show that we achieve equal or better schedulability and a faster schedule generation for most use-cases compared to simple polling approaches. Moreover, we present an extension to our method that finds the most favourable schedule for TT tasks with respect to ET schedulability, thus increasing the probability of the computed TT schedule remaining feasible when ET tasks are later added or changed.

Published
2022

41. Deconstructing the Inductive Biases of Hamiltonian Neural Networks

Author

Gruver, Nate, Finzi, Marc, Stanton, Samuel, and Wilson, Andrew Gordon

Subjects
Computer Science - Machine Learning, Mathematics - Dynamical Systems, Physics - Data Analysis, Statistics and Probability, Statistics - Machine Learning
Abstract

Physics-inspired neural networks (NNs), such as Hamiltonian or Lagrangian NNs, dramatically outperform other learned dynamics models by leveraging strong inductive biases. These models, however, are challenging to apply to many real world systems, such as those that don't conserve energy or contain contacts, a common setting for robotics and reinforcement learning. In this paper, we examine the inductive biases that make physics-inspired models successful in practice. We show that, contrary to conventional wisdom, the improved generalization of HNNs is the result of modeling acceleration directly and avoiding artificial complexity from the coordinate system, rather than symplectic structure or energy conservation. We show that by relaxing the inductive biases of these models, we can match or exceed performance on energy-conserving systems while dramatically improving performance on practical, non-conservative systems. We extend this approach to constructing transition models for common Mujoco environments, showing that our model can appropriately balance inductive biases with the flexibility required for model-based control., Comment: ICLR 2022. Code available at https://github.com/ngruver/decon-hnn

Published
2022

42. Paediatric inflammatory multisystem syndrome in Canada: population-based surveillance and role of SARS-CoV-2 linkage

Author

El Tal, Tala, Morin, Marie-Paule, Morris, Shaun K., Farrar, Daniel S., Berard, Roberta A., Kakkar, Fatima, Moore Hepburn, Charlotte, Baerg, Krista, Beaufils, Camille, Bennett, Terri-Lyn, Benseler, Susanne M., Beaudoin-Bussières, Guillaume, Chan, Kevin, Cyr, Claude, Dahdah, Nagib, Donner, Elizabeth J., Drouin, Olivier, Edjoc, Rojiemiahd, Eljaouhari, Maryem, Embree, Joanne E., Farrell, Catherine, Finzi, Andrés, Forgie, Sarah, Giroux, Ryan, Kang, Kristopher T., King, Melanie, Laffin Thibodeau, Melanie, Lang, Bianca, Laxer, Ronald M., Luu, Thuy Mai, McCrindle, Brian W., Orkin, Julia, Papenburg, Jesse, Pound, Catherine M., Price, Victoria E., Proulx-Gauthier, Jean-Philippe, Purewal, Rupeena, Sadarangani, Manish, Salvadori, Marina I., Thibeault, Roseline, Top, Karina A., Viel-Thériault, Isabelle, Haddad, Elie, Scuccimarri, Rosie, and Yeung, Rae S. M.

Published
2023
Full Text
View/download PDF

44. A Novel Method for Online Diagnostic Analysis of Partial Discharge in Instrument Transformers and Surge Arresters from the Correlation of HFCT and IEC Methods

Author

Marcel Antonionni de Andrade Romano, André Melo de Morais, Marcus Vinicius Alves Nunes, Kaynan Maresch, Luiz Fernando Freitas-Gutierres, Ghendy Cardoso, Aécio de Lima Oliveira, Erick Finzi Martins, Cristian Hans Correa, and Herber Cuadro Fontoura

Subjects
partial discharge, pulse extraction, operating condition monitoring, surge arrester, Technology
Abstract

In this work, a new methodology is proposed for the online and non-invasive extraction of partial discharge (PD) pulses from raw measurement data obtained using a simplified setup. This method enables the creation of sub-windows with optimized size, each containing a single candidate PD pulse. The proposed approach integrates mathematical morphological filtering (MMF) with kurtosis, a first-order Savitzky-Golay smoothing filter, the Otsu method for thresholding, and a specific technique to associate each sub-window with the phase angle of the applied voltage waveform, enabling the construction of phase-resolved PD (PRPD) patterns. The methodology was validated against a commercial PD detection device adhering to the IEC (International Electrotechnical Commission) standard. Experimental results demonstrated that the proposed method, utilizing an off-the-shelf 8-bit resolution data acquisition system and a low-cost high-frequency current transformer (HFCT) sensor, effectively diagnoses and characterizes PD activity in high-voltage equipment, such as surge arresters and instrument transformers, even in noisy environments. It was able to characterize PD activity using only a few cycles of the applied voltage waveform and identify low amplitude PD pulses with low signal-to-noise ratio signals. Other contribution of this work is the diagnosis and fault signature obtained from a real surge arrester (SA) with a nominal voltage of 192 kV, corroborated by destructive disassembly and internal inspection of the tested equipment. This work provides a cost-effective and accurate tool for real-time PD monitoring, which can be embedded in hardware for continuous evaluation of electrical equipment integrity.

Published
2024
Full Text
View/download PDF

Catalog

Books, media, physical & digital resources
See catalog results